Embedded AI in Smart Robots, Drones, IoT: Teaching Emerging Technologies in the Classroom

Debasis Bhattacharya, JD, DBAUniversity of Hawaii Maui College

debasisb@wawaii.eduhttp://maui.hawaii.edu/cybersecurity

Rajiv MalkanLone Star College, TX

July 25, 2019

Agenda● What is AI, Machine Learning and Deep Learning?● Embedded AI Devices – NVIDIA JETSON, NANO and AMZN Deep Lens● Teaching AI and Cybersecurity Across Disciplines● Case Studies from UH – AI Drone, Sentiment Analysis and Medical Imaging● Case Studies from Lone Star College, TX● Q&A, Discussion

2

Source: Edureka3

Source: TensorFlow for R - RStudio 4

5

Source: ACM - Computers in Entertainment - Association for Computing Machinery 6

Convolutional Neural Network - ConvNet

Source: https://ujjwalkarn.me/2016/08/11/intuitive-explanation-convnets/7

Channels - Color images have 3 channels - RGB. Each Pixel Value ranges from 0 to 255

Image - Matrix of Pixels

Source: https://ujjwalkarn.me/2016/08/11/intuitive-explanation-convnets/

Grayscale Image -One channel, values 0 (white) to 255 (black)

8

Convolution - Extract Features

3x3 Filter, or Kernel or Feature Detector

Input Image, 5x5Move 3x3 Filter over 5x5 Input Image, one pixel at a time (STRIDE) and compute matrix multiplication. Convolved Feature or Feature Map

Source: https://ujjwalkarn.me/2016/08/11/intuitive-explanation-convnets/ 9

Different values of the Filter or Kernel will create different Feature Maps from the Input Image

Examples of Features -Edges, Blurs, Sharpen

Parameters Depth - # of Filters Stride - movement across imageZero-Padding - apply filter to edges

Source: https://ujjwalkarn.me/2016/08/11/intuitive-explanation-convnets/ 10

Fully Connected Layer (FCNs)

Source: https://ujjwalkarn.me/2016/08/11/intuitive-explanation-convnets/

Rectified and Pooled Feature Maps are connected together into on FCNTraining Data Set is used to Classify Image SoftMax Activation Function used to create vector of values between 0 and 1

11

Forward Propagation

Source: https://ujjwalkarn.me/2016/08/11/intuitive-explanation-convnets/ 12

Backward Propagation Calculate Gradient of errorUse Gradient Descent to update filter weightsReduce Output Error or Training LossEpoch = Forward + Backward PropagationHyperparameter = Learning RateValidation Data -> Forward Propagation OnlyMinimize Training Loss & Validation LossControl Overfitting using DropoutsAllow for Generalization of New Test Data

Source: https://ujjwalkarn.me/2016/08/11/intuitive-explanation-convnets/ 13

Object Detection with YOLO

Source : Medium

Bounding BoxesDetect and Classify Objects

Joseph Redmon link on YOLO at Ted Talk here

14

Image Segmentation using CNN

Source: Good Audience 15

16

17

18

19

20

21

Deep Learning ToolsCNN Architectures

● LeNet● AlexNet● ZF Net● GoogLeNet● ResNets

Frameworks

● TensorFlow● Caffe● MS CNTK● PyTorch/Torch● MXNet● Chainer● Keras● Theano● Deeplearning4j● PaddlePaddle● MATLAB

● cuDNN● DIGITS● DetectNet● TensorRT● RAPIDS● JETSON 22

Teaching AI and Cybersecurity Across Disciplines● Electronics● Healthcare● Hospitality and Tourism● Business, Finance, Accounting● Criminal Justice● Computer Science● Mathematics ● Etc.

● How do you teach the essence of AI across the disciplines?

23

Case Study: Sentiment Analysis in Hospitality/Tourism

● Sentiment Analysis may be performed as an application of Machine Learning (ML) to large bodies of text, such as those found in large consumer review datasets, in order to determine sentiment (positive, negative, sarcastic, etc.) and gain feedback.

● The use of Machine Learning techniques in this endeavor allows for much larger quantities of data to be processed than would be practical for human evaluators working directly with the data.

● With recent advances in Machine Learning in the form of new and powerful frameworks, it is relatively simple to set up a machine to perform analysis on text in a way previously confined to the domain of commonsense, human interpretation of opinions, feelings, etc.

● Sample Code - https://github.com/UHMC/nifty-sentiment-analysis 24

Case Study: Neural Network for Medical Image Processing● Medical imaging is becoming an increasingly popular application of Machine

Learning. Medical practitioners can use software to obtain diagnosis or second opinions on X-Ray images, lowering the chances of a missed threat.

● In this assignment, students will be able to set up a model to train using the Deep Learning Tool Kit and Tensorflow.

● Applicable for Students in various disciplines – CompSci, Healthcare etc.● Source Code - https://github.com/UHMC/nifty-medical-imaging

25

http://www.lonestar.edu/land-survey-aas.htm

Best Practices from Lone Star College, TX

26

27

28

29

30

31

32

Questions? CommentsDebasis Bhattacharya, JD, DBA

University of Hawaii Maui Collegedebasisb@wawaii.edu

http://maui.hawaii.edu/cybersecurityRajiv Malkan

Lone Star College, TXJuly 25, 2019

33